Motion Estimation for Perceptual Image Sequence Coding

Since the advent of television obtaining high perceived quality using a limited bandwidth has been an important issue. This thesis proposes new methods for exploitation of temporal and perceptual redundancy in image sequences to achieve lower coding rate and/or higher visual quality. The methods presented are inspired and based on human visual system models. Particularly relevant in the present context are the indications that the visual cortex contains cells that are selective in orientation and frequency but invariant to the phase of the stimuli. For this reason a spatial quadrature filter bank, representing images in a similar fashion, is generated. For computational efficiency, a filter net technique is employed using combinations of simple sequential 1D filter kernels. The filter bank is designed for interlaced video which is still the most common format for video sequences. For coding of image sequences temporal redundancy is reduced using motion compensated prediction. In motion compensated prediction the prediction of the next image is given by the present image and a predicted dense local motion field. Motion compensation is performed with a new and computationally efficient method. The method estimates data samples on a desired output grid from input data represented by samples on an irregularly grid. The initially predicted image is refined using forward motion compensation with a sparse motion field. In this case only the sparse motion field needs to be transmitted to the decoder. As a result a prediction without block artifacts, common in standard forward motion compensation schemes, is generated. Experiments show that this method performs better than traditional block-matching approaches. The motion is estimated using a new approach based on phase differences computed from products of quadrature filter responses. The approach includes learning parameters for motion estimation and introduces multiple hierarchical motion estimation to achieve estimates with high spatial resolution. The quadrature filter bank approach used for motion estimation also provides a basis for image quality estimation in accordance with human perception. This allows the video quality estimator to be an integral part of the video coder and opens up the possibility of local space-time optimization of video coding parameters.